github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-06-28 16:41:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

Discard HAVE_AVX512VBMI checks at Sheng/McSheng compile time.

Browse Source
This commit is contained in:
Chang, Harry 2020-10-21 12:30:04 +00:00 committed by Konstantinos Margaritis
parent b19a41528a
commit 5ad3d64b4b
12 changed files with 18 additions and 77 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/nfa/mcsheng.h
View File

@ -152,6 +152,6 @@ char nfaExecMcSheng64_16_expandState(const struct NFA *nfa, void *dest,
#define nfaExecMcSheng64_16_testEOD NFA_API_NO_IMPL
#define nfaExecMcSheng64_16_reportCurrent NFA_API_NO_IMPL
#endif //end of HAVE_AVX512VBM
#endif //end of HAVE_AVX512VBMI
#endif

15
src/nfa/mcsheng_compile.cpp
View File

@ -243,7 +243,6 @@ void populateBasicInfo(size_t state_size, const dfa_info &info,
}
}
#if defined(HAVE_AVX512VBMI)
static
mstate_aux *getAux64(NFA *n, dstate_id_t i) {
mcsheng64 *m = (mcsheng64 *)getMutableImplNfa(n);
@ -343,7 +342,6 @@ void populateBasicInfo64(size_t state_size, const dfa_info &info,
m->flags |= MCSHENG_FLAG_SINGLE;
}
}
#endif
static
size_t calcShermanRegionSize(const dfa_info &info) {
@ -719,7 +717,6 @@ void fill_in_succ_table_16(NFA *nfa, const dfa_info &info,
}
}
#if defined(HAVE_AVX512VBMI)
static
void fill_in_aux_info64(NFA *nfa, const dfa_info &info,
const map<dstate_id_t, AccelScheme> &accel_escape_info,
@ -793,7 +790,6 @@ void fill_in_succ_table_64_16(NFA *nfa, const dfa_info &info,
}
}
}
#endif
#define MAX_SHERMAN_LIST_LEN 8
@ -1113,7 +1109,6 @@ void fill_in_succ_table_8(NFA *nfa, const dfa_info &info,
}
}
#if defined(HAVE_AVX512VBMI)
static
void fill_in_sherman64(NFA *nfa, dfa_info &info, UNUSED u16 sherman_limit) {
char *nfa_base = (char *)nfa;
@ -1267,7 +1262,6 @@ void fill_in_succ_table_64_8(NFA *nfa, const dfa_info &info,
}
}
}
#endif
static
void allocateImplId8(dfa_info &info, dstate_id_t sheng_end,
@ -1366,7 +1360,6 @@ bytecode_ptr<NFA> mcshengCompile8(dfa_info &info, dstate_id_t sheng_end,
return nfa;
}
#if defined(HAVE_AVX512VBMI)
static
bytecode_ptr<NFA> mcsheng64Compile8(dfa_info &info, dstate_id_t sheng_end,
const map<dstate_id_t, AccelScheme> &accel_escape_info) {
@ -1418,7 +1411,6 @@ bytecode_ptr<NFA> mcsheng64Compile8(dfa_info &info, dstate_id_t sheng_end,
return nfa;
}
#endif
bytecode_ptr<NFA> mcshengCompile(raw_dfa &raw, const CompileContext &cc,
const ReportManager &rm) {
@ -1468,13 +1460,17 @@ bytecode_ptr<NFA> mcshengCompile(raw_dfa &raw, const CompileContext &cc,
return nfa;
}
#if defined(HAVE_AVX512VBMI)
bytecode_ptr<NFA> mcshengCompile64(raw_dfa &raw, const CompileContext &cc,
const ReportManager &rm) {
if (!cc.grey.allowMcSheng) {
return nullptr;
}
if (!cc.target_info.has_avx512vbmi()) {
DEBUG_PRINTF("McSheng64 failed, no HS_CPU_FEATURES_AVX512VBMI!\n");
return nullptr;
}
mcclellan_build_strat mbs(raw, rm, false);
dfa_info info(mbs);
bool using8bit = cc.grey.allowMcClellan8 && info.size() <= 256;
@ -1523,7 +1519,6 @@ bytecode_ptr<NFA> mcshengCompile64(raw_dfa &raw, const CompileContext &cc,
DEBUG_PRINTF("compile done\n");
return nfa;
}
#endif
bool has_accel_mcsheng(const NFA *) {
return true; /* consider the sheng region as accelerated */

1
src/nfa/mcsheng_data.c
View File

@ -53,4 +53,3 @@ const u64a mcsheng64_pext_mask[8] = {
0xff0000000000003f,
};
#endif

10
src/nfa/mcsheng_dump.cpp
View File

@ -174,7 +174,6 @@ void describeEdge(FILE *f, const mcsheng *m, const u16 *t, u16 i) {
}
}
#if defined(HAVE_AVX512VBMI)
static
const mstate_aux *getAux64(const NFA *n, dstate_id_t i) {
auto *m = (const mcsheng64 *)getImplNfa(n);
@ -292,7 +291,6 @@ void describeEdge64(FILE *f, const mcsheng64 *m, const u16 *t, u16 i) {
fprintf(f, "\" ];\n");
}
}
#endif
static
void dumpAccelDot(FILE *f, u16 i, const union AccelAux *accel) {
@ -376,7 +374,6 @@ void describeNode(const NFA *n, const mcsheng *m, u16 i, FILE *f) {
}
#if defined(HAVE_AVX512VBMI)
static
void describeNode64(const NFA *n, const mcsheng64 *m, u16 i, FILE *f) {
const mstate_aux *aux = getAux64(n, i);
@ -436,7 +433,6 @@ void describeNode64(const NFA *n, const mcsheng64 *m, u16 i, FILE *f) {
}
}
#endif
static
void dumpDotPreambleDfa(FILE *f) {
@ -574,7 +570,6 @@ void dump_text_8(const NFA *nfa, FILE *f) {
dumpTextReverse(nfa, f);
}
#if defined(HAVE_AVX512VBMI)
static
void dump64_dot_16(const NFA *nfa, FILE *f) {
auto *m = (const mcsheng64 *)getImplNfa(nfa);
@ -699,7 +694,6 @@ void dump64_text_16(const NFA *nfa, FILE *f) {
fprintf(f, "\n");
dumpTextReverse(nfa, f);
}
#endif
void nfaExecMcSheng16_dump(const NFA *nfa, const string &base) {
assert(nfa->type == MCSHENG_NFA_16);
@ -714,19 +708,15 @@ void nfaExecMcSheng8_dump(const NFA *nfa, const string &base) {
}
void nfaExecMcSheng64_16_dump(UNUSED const NFA *nfa, UNUSED const string &base) {
#if defined(HAVE_AVX512VBMI)
assert(nfa->type == MCSHENG_64_NFA_16);
dump64_text_16(nfa, StdioFile(base + ".txt", "w"));
dump64_dot_16(nfa, StdioFile(base + ".dot", "w"));
#endif
}
void nfaExecMcSheng64_8_dump(UNUSED const NFA *nfa, UNUSED const string &base) {
#if defined(HAVE_AVX512VBMI)
assert(nfa->type == MCSHENG_64_NFA_8);
dump64_text_8(nfa, StdioFile(base + ".txt", "w"));
dump64_dot_8(nfa, StdioFile(base + ".dot", "w"));
#endif
}
} // namespace ue2

2
src/nfa/mcsheng_internal.h
View File

@ -92,7 +92,6 @@ struct mcsheng {
* representing the data from a u64a. */
extern const u64a mcsheng_pext_mask[8];
#if defined(HAVE_AVX512VBMI)
struct mcsheng64 {
u16 state_count; /**< total number of states */
u32 length; /**< length of dfa in bytes */
@ -121,6 +120,5 @@ struct mcsheng64 {
};
extern const u64a mcsheng64_pext_mask[8];
#endif
#endif

14
src/nfa/nfa_internal.h
View File

@ -152,12 +152,8 @@ static really_inline int isMcClellanType(u8 t) {
/** \brief True if the given type (from NFA::type) is a Sheng-McClellan hybrid
* DFA. */
static really_inline int isShengMcClellanType(u8 t) {
#if defined(HAVE_AVX512VBMI)
return t == MCSHENG_64_NFA_8 || t == MCSHENG_64_NFA_16 || t == MCSHENG_NFA_8 ||
t == MCSHENG_NFA_16;
#else
return t == MCSHENG_NFA_8 || t == MCSHENG_NFA_16;
#endif
return t == MCSHENG_NFA_8 || t == MCSHENG_NFA_16 ||
t == MCSHENG_64_NFA_8 || t == MCSHENG_64_NFA_16;
}
/** \brief True if the given type (from NFA::type) is a Gough DFA. */
@ -170,7 +166,6 @@ static really_inline int isSheng16Type(u8 t) {
return t == SHENG_NFA;
}
#if defined(HAVE_AVX512VBMI)
/** \brief True if the given type (from NFA::type) is a Sheng32 DFA. */
static really_inline int isSheng32Type(u8 t) {
return t == SHENG_NFA_32;
@ -180,15 +175,10 @@ static really_inline int isSheng32Type(u8 t) {
static really_inline int isSheng64Type(u8 t) {
return t == SHENG_NFA_64;
}
#endif
/** \brief True if the given type (from NFA::type) is a Sheng16/32/64 DFA. */
static really_inline int isShengType(u8 t) {
#if defined(HAVE_AVX512VBMI)
return t == SHENG_NFA || t == SHENG_NFA_32 || t == SHENG_NFA_64;
#else
return t == SHENG_NFA;
#endif
}
/**

4
src/nfa/sheng_internal.h
View File

@ -38,7 +38,6 @@
#define SHENG_STATE_MASK 0xF
#define SHENG_STATE_FLAG_MASK 0x70
#if defined (HAVE_AVX512VBMI)
#define SHENG32_STATE_ACCEPT 0x20
#define SHENG32_STATE_DEAD 0x40
#define SHENG32_STATE_ACCEL 0x80
@ -49,7 +48,6 @@
#define SHENG64_STATE_DEAD 0x80
#define SHENG64_STATE_MASK 0x3F
#define SHENG64_STATE_FLAG_MASK 0xC0
#endif
#define SHENG_FLAG_SINGLE_REPORT 0x1
#define SHENG_FLAG_CAN_DIE 0x2
@ -80,7 +78,6 @@ struct sheng {
ReportID report;
};
#if defined (HAVE_AVX512VBMI)
struct sheng32 {
m512 succ_masks[256];
u32 length;
@ -106,6 +103,5 @@ struct sheng64 {
u8 flags;
ReportID report;
};
#endif
#endif /* SHENG_INTERNAL_H_ */

20
src/nfa/shengcompile.cpp
View File

@ -302,7 +302,6 @@ void dumpShuffleMask(const u8 chr, const u8 *buf, unsigned sz) {
DEBUG_PRINTF("chr %3u: %s\n", chr, o.str().c_str());
}
#if defined (HAVE_AVX512VBMI)
static really_inline
void dumpShuffleMask32(const u8 chr, const u8 *buf, unsigned sz) {
stringstream o;
@ -325,7 +324,6 @@ void dumpShuffleMask64(const u8 chr, const u8 *buf, unsigned sz) {
DEBUG_PRINTF("chr %3u: %s\n", chr, o.str().c_str());
}
#endif
#endif
static
void fillAccelOut(const map<dstate_id_t, AccelScheme> &accel_escape_info,
@ -358,7 +356,6 @@ u8 getShengState<sheng>(dstate &state, dfa_info &info,
return s;
}
#if defined(HAVE_AVX512VBMI)
template <>
u8 getShengState<sheng32>(dstate &state, dfa_info &info,
map<dstate_id_t, AccelScheme> &accelInfo) {
@ -387,7 +384,6 @@ u8 getShengState<sheng64>(dstate &state, dfa_info &info,
}
return s;
}
#endif
template <typename T>
static
@ -446,7 +442,6 @@ void populateBasicInfo<sheng>(struct NFA *n, dfa_info &info,
s->floating = getShengState<sheng>(info.floating, info, accelInfo);
}
#if defined(HAVE_AVX512VBMI)
template <>
void populateBasicInfo<sheng32>(struct NFA *n, dfa_info &info,
map<dstate_id_t, AccelScheme> &accelInfo,
@ -496,7 +491,6 @@ void populateBasicInfo<sheng64>(struct NFA *n, dfa_info &info,
s->anchored = getShengState<sheng64>(info.anchored, info, accelInfo);
s->floating = getShengState<sheng64>(info.floating, info, accelInfo);
}
#endif
template <typename T>
static
@ -582,7 +576,6 @@ bool createShuffleMasks<sheng>(sheng *s, dfa_info &info,
return true;
}
#if defined(HAVE_AVX512VBMI)
template <>
bool createShuffleMasks<sheng32>(sheng32 *s, dfa_info &info,
map<dstate_id_t, AccelScheme> &accelInfo) {
@ -627,7 +620,6 @@ bool createShuffleMasks<sheng64>(sheng64 *s, dfa_info &info,
}
return true;
}
#endif
bool has_accel_sheng(const NFA *) {
return true; /* consider the sheng region as accelerated */
@ -731,7 +723,6 @@ bytecode_ptr<NFA> shengCompile(raw_dfa &raw, const CompileContext &cc,
return shengCompile_int<sheng>(raw, cc, accel_states, strat, info);
}
#if defined(HAVE_AVX512VBMI)
bytecode_ptr<NFA> sheng32Compile(raw_dfa &raw, const CompileContext &cc,
const ReportManager &rm, bool only_accel_init,
set<dstate_id_t> *accel_states) {
@ -740,6 +731,11 @@ bytecode_ptr<NFA> sheng32Compile(raw_dfa &raw, const CompileContext &cc,
return nullptr;
}
if (!cc.target_info.has_avx512vbmi()) {
DEBUG_PRINTF("Sheng32 failed, no HS_CPU_FEATURES_AVX512VBMI!\n");
return nullptr;
}
sheng_build_strat strat(raw, rm, only_accel_init);
dfa_info info(strat);
@ -767,6 +763,11 @@ bytecode_ptr<NFA> sheng64Compile(raw_dfa &raw, const CompileContext &cc,
return nullptr;
}
if (!cc.target_info.has_avx512vbmi()) {
DEBUG_PRINTF("Sheng64 failed, no HS_CPU_FEATURES_AVX512VBMI!\n");
return nullptr;
}
sheng_build_strat strat(raw, rm, only_accel_init);
dfa_info info(strat);
@ -790,6 +791,5 @@ bytecode_ptr<NFA> sheng64Compile(raw_dfa &raw, const CompileContext &cc,
}
return nfa;
}
#endif
} // namespace ue2

2
src/nfa/shengcompile.h
View File

@ -71,7 +71,6 @@ bytecode_ptr<NFA> shengCompile(raw_dfa &raw, const CompileContext &cc,
const ReportManager &rm, bool only_accel_init,
std::set<dstate_id_t> *accel_states = nullptr);
#if defined(HAVE_AVX512VBMI)
bytecode_ptr<NFA> sheng32Compile(raw_dfa &raw, const CompileContext &cc,
const ReportManager &rm, bool only_accel_init,
std::set<dstate_id_t> *accel_states = nullptr);
@ -79,7 +78,6 @@ bytecode_ptr<NFA> sheng32Compile(raw_dfa &raw, const CompileContext &cc,
bytecode_ptr<NFA> sheng64Compile(raw_dfa &raw, const CompileContext &cc,
const ReportManager &rm, bool only_accel_init,
std::set<dstate_id_t> *accel_states = nullptr);
#endif
struct sheng_escape_info {
CharReach outs;

20
src/nfa/shengdump.cpp
View File

@ -64,7 +64,6 @@ const sstate_aux *get_aux(const NFA *n, dstate_id_t i) {
return aux;
}
#if defined(HAVE_AVX512VBMI)
static
const sstate_aux *get_aux32(const NFA *n, dstate_id_t i) {
assert(n && isSheng32Type(n->type));
@ -94,7 +93,6 @@ const sstate_aux *get_aux64(const NFA *n, dstate_id_t i) {
return aux;
}
#endif
static
void dumpHeader(FILE *f, const sheng *s) {
@ -111,7 +109,6 @@ void dumpHeader(FILE *f, const sheng *s) {
!!(s->flags & SHENG_FLAG_SINGLE_REPORT));
}
#if defined(HAVE_AVX512VBMI)
static
void dumpHeader32(FILE *f, const sheng32 *s) {
fprintf(f, "number of states: %u, DFA engine size: %u\n", s->n_states,
@ -141,7 +138,6 @@ void dumpHeader64(FILE *f, const sheng64 *s) {
!!(s->flags & SHENG_FLAG_CAN_DIE),
!!(s->flags & SHENG_FLAG_SINGLE_REPORT));
}
#endif
static
void dumpAux(FILE *f, u32 state, const sstate_aux *aux) {
@ -151,7 +147,6 @@ void dumpAux(FILE *f, u32 state, const sstate_aux *aux) {
aux->top & SHENG_STATE_MASK);
}
#if defined(HAVE_AVX512VBMI)
static
void dumpAux32(FILE *f, u32 state, const sstate_aux *aux) {
fprintf(f, "state id: %u, reports offset: %u, EOD reports offset: %u, "
@ -167,7 +162,6 @@ void dumpAux64(FILE *f, u32 state, const sstate_aux *aux) {
state, aux->accept, aux->accept_eod, aux->accel,
aux->top & SHENG64_STATE_MASK);
}
#endif
static
void dumpReports(FILE *f, const report_list *rl) {
@ -197,7 +191,6 @@ void dumpMasks(FILE *f, const sheng *s) {
}
}
#if defined(HAVE_AVX512VBMI)
static
void dumpMasks32(FILE *f, const sheng32 *s) {
for (u32 chr = 0; chr < 256; chr++) {
@ -237,7 +230,6 @@ void dumpMasks64(FILE *f, const sheng64 *s) {
fprintf(f, "\n");
}
}
#endif
static
void nfaExecSheng_dumpText(const NFA *nfa, FILE *f) {
@ -277,7 +269,6 @@ void nfaExecSheng_dumpText(const NFA *nfa, FILE *f) {
fprintf(f, "\n");
}
#if defined(HAVE_AVX512VBMI)
static
void nfaExecSheng32_dumpText(const NFA *nfa, FILE *f) {
assert(nfa->type == SHENG_NFA_32);
@ -353,7 +344,6 @@ void nfaExecSheng64_dumpText(const NFA *nfa, FILE *f) {
fprintf(f, "\n");
}
#endif
static
void dumpDotPreambleDfa(FILE *f) {
@ -401,7 +391,6 @@ void describeNode<sheng>(const NFA *n, const sheng *s, u16 i, FILE *f) {
}
}
#if defined(HAVE_AVX512VBMI)
template <>
void describeNode<sheng32>(const NFA *n, const sheng32 *s, u16 i, FILE *f) {
const sstate_aux *aux = get_aux32(n, i);
@ -461,7 +450,6 @@ void describeNode<sheng64>(const NFA *n, const sheng64 *s, u16 i, FILE *f) {
fprintf(f, "STARTF -> %u [color = red ]\n", i);
}
}
#endif
static
void describeEdge(FILE *f, const u16 *t, u16 i) {
@ -514,7 +502,6 @@ void shengGetTransitions(const NFA *n, u16 state, u16 *t) {
t[TOP] = aux->top & SHENG_STATE_MASK;
}
#if defined(HAVE_AVX512VBMI)
static
void sheng32GetTransitions(const NFA *n, u16 state, u16 *t) {
assert(isSheng32Type(n->type));
@ -550,7 +537,6 @@ void sheng64GetTransitions(const NFA *n, u16 state, u16 *t) {
t[TOP] = aux->top & SHENG64_STATE_MASK;
}
#endif
static
void nfaExecSheng_dumpDot(const NFA *nfa, FILE *f) {
@ -572,7 +558,6 @@ void nfaExecSheng_dumpDot(const NFA *nfa, FILE *f) {
fprintf(f, "}\n");
}
#if defined(HAVE_AVX512VBMI)
static
void nfaExecSheng32_dumpDot(const NFA *nfa, FILE *f) {
assert(nfa->type == SHENG_NFA_32);
@ -612,7 +597,6 @@ void nfaExecSheng64_dumpDot(const NFA *nfa, FILE *f) {
fprintf(f, "}\n");
}
#endif
void nfaExecSheng_dump(const NFA *nfa, const string &base) {
assert(nfa->type == SHENG_NFA);
@ -621,19 +605,15 @@ void nfaExecSheng_dump(const NFA *nfa, const string &base) {
}
void nfaExecSheng32_dump(UNUSED const NFA *nfa, UNUSED const string &base) {
#if defined(HAVE_AVX512VBMI)
assert(nfa->type == SHENG_NFA_32);
nfaExecSheng32_dumpText(nfa, StdioFile(base + ".txt", "w"));
nfaExecSheng32_dumpDot(nfa, StdioFile(base + ".dot", "w"));
#endif
}
void nfaExecSheng64_dump(UNUSED const NFA *nfa, UNUSED const string &base) {
#if defined(HAVE_AVX512VBMI)
assert(nfa->type == SHENG_NFA_64);
nfaExecSheng64_dumpText(nfa, StdioFile(base + ".txt", "w"));
nfaExecSheng64_dumpDot(nfa, StdioFile(base + ".dot", "w"));
#endif
}
} // namespace ue2

3
src/rose/rose_build_bytecode.cpp
View File

@ -632,8 +632,6 @@ bytecode_ptr<NFA> getDfa(raw_dfa &rdfa, bool is_transient,
* bytecode and that they are usually run on small blocks */
dfa = mcshengCompile(rdfa, cc, rm);
}
#if defined(HAVE_AVX512VBMI)
if (!dfa) {
dfa = sheng32Compile(rdfa, cc, rm, false);
}
@ -643,7 +641,6 @@ bytecode_ptr<NFA> getDfa(raw_dfa &rdfa, bool is_transient,
if (!dfa && !is_transient) {
dfa = mcshengCompile64(rdfa, cc, rm);
}
#endif
if (!dfa) {
// Sheng wasn't successful, so unleash McClellan!
dfa = mcclellanCompile(rdfa, cc, rm, false);

2
src/smallwrite/smallwrite_build.cpp
View File

@ -793,14 +793,12 @@ bytecode_ptr<NFA> getDfa(raw_dfa &rdfa, const CompileContext &cc,
bytecode_ptr<NFA> dfa = nullptr;
if (cc.grey.allowSmallWriteSheng) {
dfa = shengCompile(rdfa, cc, rm, only_accel_init, &accel_states);
#if defined(HAVE_AVX512VBMI)
if (!dfa) {
dfa = sheng32Compile(rdfa, cc, rm, only_accel_init, &accel_states);
}
if (!dfa) {
dfa = sheng64Compile(rdfa, cc, rm, only_accel_init, &accel_states);
}
#endif
}
if (!dfa) {
dfa = mcclellanCompile(rdfa, cc, rm, only_accel_init,